[EM] Burlington Vermont repeals IRV 52% to 48%

[Kristofer Munsterhjelm]

robert bristow-johnson wrote:
> 
> On Mar 6, 2010, at 2:34 PM, Raph Frank wrote:
> 
>> On Sat, Mar 6, 2010 at 1:34 AM, robert bristow-johnson
>> <rbj at audioimagination.com> wrote:
> 
>>> i almost never vote for all 6.  usually just 2.  but it's a strategic 
>>> vote.  and since i didn't hit the limit, it's practically no 
>>> different than Approval voting.  i cannot see how Warren and company 
>>> claim that it's less strategic than Condorcet.
>>
>> Assuming 6 seats, optimal strategy is probably to approve either the
>> 6th or 7th most popular candidate and all candidates you prefer to
>> that candidate.  If you run out of votes, the you should approve
>> candidates who are most likely to tie for the 6th seat first before
>> other candidates.
>>
>> In any case, you really should cast all 6 votes.
> 
> i don't get it.  just because the party i most identify with proffers 6 
> candidates (as does two or three other parties) doesn't mean that i, as 
> a independently-minded voter, care if all of those candidates are 
> elected.  if i "approve" of *all* of those candidates, it's only because 
> of blind party affiliation.

It's strategy. Approval voting is really just a name, and when you get 
into strategy, I don't think it matches the common idea of approval 
because you have to set the threshold depending on poll data.

In any case, the 6 candidates don't all have to be from the same party; 
and I don't know what strategy would be used for a multiwinner Approval 
system, since none that I know are any good (at least not according to 
my simulations).

> i know that if my candidate *is* elected, 5 other candidates will also 
> be elected but i have much less interest in who those 5 will be.  or, i 
> might recognize some of those candidates to be shoe-ins and that they'll 
> likely be elected from the support they are receiving elsewhere.  so, if 
> that were the political interests i am bringing to the polls, why would 
> i choose to harm the chances of this candidate i really like, by 
> approving any other candidate?
> 
> i actually think that, even in a multi-winner election, that Condorcet 
> ordering of the candidates could make sense (with the top 6 preferred 
> candidates elected).  of course there is a problem if there is a cycle 
> that is transected by the cutoff boundary of the top 6 who get elected 
> and those lower who do not.  i am not seriously proposing actually 
> implementing this without some serious study, and a good method (perhaps 
> Ranked Pairs or Schulze) would be needed to deal with a cycle that 
> crosses the win/lose boundary.

Straightforwardly porting single-winner methods to multi-winner cases 
don't really work. A good single-winner method finds acceptable 
compromises, centrists that are good in the eyes of all, subject to 
majority rule. However, for multiwinner methods, one usually desires 
proportionality in the respects that matter instead, so each group (be 
that a group of people, or of opinions, etc) can be represented.

One concrete example is this:
51: A1 > A2 > A3 > B1 > B2 > B3
49: B1 > B2 > B3 > A1 > A2 > A3

Majority rule means we should have A1 in the single-winner case. A 
Condorcet method will output a rank corresponding to the 51% majority 
(that is, A1 > A2 > A3 > B1 > B2 > B3). A council of two would thus have 
two A-candidates, but if we desire proportionality, there should be one 
of each - A1 and B1.

> that said, Approval voting requires more strategy from me than just 
> ranking candidates in my preferred order.  whether it's a single or 
> multi-winner election, i really think that the ranked ballot is the 
> simplest way to extract necessary information from voters, without 
> expecting too much from voters (which is what Range or Score voting does).

If people absolutely want score ballots (or ranked plus cutoff or 
whatever), then they may have it, but generally, I agree here. The 
greater problem of Approval is (IMHO) that it forces a strategy. Not 
only do the voters have to decide between "want" and "want not", they 
have to dynamically adjust the threshold. Not only do Range voters have 
to find out how much they like a certain option, if they want their 
votes to count, they have to rescale the values.

It is possible to use a rated ballot in a Condorcet manner (e.g. 
Cardinal Weighted Pairwise) if people really want to have such a ballot. 
Such a method wouldn't have the disadvantages of Range. Of course, it's 
a tradeoff, as the method no longer passes FBC, for instance.